Crazy coincidence, I beat this game on Switch and then watched the TAS just a few days ago. There is some surprisingly great manipulation of enemies and hitboxes in it, but mostly just normal optimized solutions, so probably still Vault.
Congratulations on finding an improvement!
I realized Left+Right can be used to input Left on two consecutive frames. This is (hopefully) saving frames in both TCG1 runs, but it also has one usage in this run (when inputting "C"), saving 1 frame in the deck naming.
Unfortunately, on the interim build (which is console accurate), saving this one frame in the deck edit adds 4 lag frames to the duel. There's not much to work with in this run for lag reduction, and since it's only a 1 frame input improvement, I think just documenting it here is best.
For context in this thread: the Zool submission has brought up questions about accepting 'trivial' runs that beat existing movies: http://tasvideos.org/forum/viewtopic.php?p=492783#492783
I'm really not sure about a 3-input glitched run obsoleting a full-game run here. In Zool's case it was a glitch execution being improved to the point that it's short enough to grind for in real-time. The published 3 minute version basically no longer exists, it's not a record of any kind.
But a full-game run of this does still exist. It seems like there should be a middle ground between "accept a glitch improvement if it retroactively becomes trivial" and "accept new 'trivial' glitches".
And from the sound of the description, this TAS is a recreation of an RTA run, as opposed to the RTA runs duplicating the TAS like Dragster/Zool (which should not necessarily be penalized).
I am very biased in enjoying the movie, as B-K is my favorite game, and the first game I ever saw a speedrun of. But I definitely think it deserves Star tier. (I mentioned that in a chat and EZGames told me to post here...)
To me Star tier is about movies that someone that's never watched a TAS before would really like, and that tends to mean extremely high-effort runs of popular games. I do think Banjo qualifies as both. Banjo's recent inclusion in Smash has shown that he still has a LOT of love.
As for Kazooie's "repeating sounds", I could spend hours listening to them in the background... And with recent finds (810 note door skip, and Click Clock Wood early), an any% run's main limiting collectible would be... 45 Mumbo Tokens to transform into Pumpkin and Bee, I believe. It would not be more entertaining.
Would that Pinocchio run be accepted then since the game was accepted before? Rather than "acceptability of games, not movies", it may make more sense for it to be "acceptability of categories". New game end glitches can definitely retroactively make a fastest-completion trivial.
This movie, well, I would say it's not entertaining. Particularly because there's no music during the credits...
2. There is a slower movie published, so publishing the maxed out time is important to Vault's mission of record-keeping. If there wasn't a previous movie, rejecting would make more sense, but as it is, the game has been maxed and deserves to be recorded as such
What about the Pinocchio submission? Should that be accepted too, even if it's extremely trivial?
I don't think Pinocchio is the same situation since it's comparing a full game run to a maxed "game end glitch" run. Though Vault rules don't generally care about that difference, one can say that the full game run is still the fastest completion that follows submission rules, and therefore there's no need to accept the trivial one for record-keeping.
This is comparing a non-maxed "game end glitch" to a now properly maxed "game end glitch" run, so since it's the same glitch, the non-maxed publication is now no longer beating all (comparable) records.
I have a couple main thoughts on why rejection seems wrong that I don't think came up here
1. The time was first achieved in TAS, and later duplicated in real time thanks to the frame timings needed being discovered through TASing; so though it wasn't submitted yet (due to 8bit thoroughly brute forcing every possibility!), Zool in 21.61 beat all records at time of creation. (same situation as maxing Dragster's in-game time)
2. There is a slower movie published, so publishing the maxed out time is important to Vault's mission of record-keeping. If there wasn't a previous movie, rejecting would make more sense, but as it is, the game has been maxed and deserves to be recorded as such
In the end it doesn't reeeeally matter that much, but, I suppose the precedent is more important than the 8 second run. Though generally, "maxed" times will probably be this short.
Fantastic TAS, and immensely great explanation video. I've followed DK64 runs off and on for years, and to get such a thorough explanation of the WHY behind tricks like swim through vertical walls (which now also includes walk through walls!), moonkicks, Main Menu Moves and telegrabbing was great.
What little requirements are left in this game's any%, and the Intro Story timing and boss segments, come together to have in my opinion a perfect balance between quick progression and playaround time. Managing to find time to show K. Rool's control room is the peak of entertainment to me! David Wonn would be so proud!
The new glitch is cool, but the walking/scroll speed isn't too exciting, and the long stretches of flying above the map, well... In stage 4 I skipped ahead a bit...
Just posting this here for possible future reference
I began work on updated TCG1 TAS, and thoroughly checked the starting name used, something that hadn't been checked before.
Oddly, though they all take the same number of frames to input, "No" (as was chosen for humor purposes) is fastest in the TCG1 ACE TAS, with the other 9 options having extra lag frames the first time your hand is displayed in the tutorial. ("R" got 2 extra, the rest had 4 extra.)
I wanted to confirm there was no possible time save based on name choice in this game as well. My options are drastically more limited, as name entry in this game is 3 frames instead of 4. (In TCG1, we had to wait for A to be de-pressed)
Default character (A, Start, A) had 3 extra lag frames displaying the Grass Energy text in duel.
The other three options (Right/Down/Left+A, Start, A) all have identical end.
As TCG1's No saved more than 1 frame, I decided to try 4-frame names in TCG2 as well. They all had identical Grass Energy lag length, but lost 2 extra lag frames before the duel, so they came out 3 frames slower (and they appeared to sync regardless of name).
To be extra thorough, I checked choosing Boy as gender. Oddly, Right+A then was 2 lag frames slower. Left+A and Down+A remained the same that's submitted, and A was still 3 slower. I also tried different valid inputs like "Left+A, Up, A" or "Left+A, x, Up+A" (since Left wraps you around to be an Up away from Start), and they all came out the same.
tl;dr: Name and gender CAN matter, but as far as I tested, I happened to choose one that doesn't lose time.
Fraplatek has been improving their time, which is very cool:
Link to video
They've found an 8-line clear RNG here, though with a pretty weak garbage layout.
I'd like to try it, if I got a movie file or could duplicate the RNG.
I can't think of any way to really effectively automate finding RNGs in Lua, I'm thinking about trying to make some kind of brute forcer that has tables defining different useless inputs (like P1 Up + P2 Down) and "go up" inputs, though it'd be hardish to account for making sure to not press the same input 2 frames in a row.
Hmm, I would agree, the 2:40 of intro time in TCG1 is more entertaining than the 1:40 of intro time in TCG2, since you can read it.
The RNG manip and clock cycle manip required in TCG1 is also more impressive on a technical level. I have a 1:59 RTA time in TCG2 with this strategy, since there's no RNG involved and it always succeeds. So it's less of a technical achievement as well, and really just shows TAS levels of keyboard navigation compared to RTA. (Though I do think CasualPokePlayer's idea of placing an E9 inside D3xx was a genius work around to deck naming not having direct access to any Jump opcodes)
Here's an interesting question:
What do you think of a new version of the full game TAS which uses Duel Escape for the tutorial, but not for any other duels? It's technically a speed/entertainment trade-off, as having to carefully manipulate RNG to embarrass every opponent in one turn is more entertaining than pressing Down+A.
I believe the tutorial time is the main thing killing its Mooniness. Although the menuing is still perhaps too fast to 'follow' as well.
And with the major menuing improvement gifvex found for up/down, there will be optimization to do throughout the movie's deck edits and duel input as well, so it wouldn't just be a copy/paste job. For example, the down+L/R menuing is so fast, the interim card pictures can't even load, saving lag frames too (DCE picture for using PlusPower):
I managed 28 frames save on the movie. I believe it's:
8 frames from starting quick drops earlier
1 frame from placing the useless Z at the end one row earlier by rotating it
19 frames from rotation pushes (It's an odd number, because for one of them, I used a 1-frame Pause to be able to press B two game frames in a row, to avoid an early piece lock)
I think a different board setup would be needed to get more pushes than this, but everything I tried for improving possible pushes, just makes the board not quite fill in enough to survive. So, I'm once again happy with where it is.
Spikestuff made a comparison video:
Link to video
Here's the movie file to update the submission: http://tasvideos.org/userfiles/info/61562979381368213
Oopsies, I found an easy 8 frame improvement. I didn't realize that rotations can be done while quick dropping without resetting the quick drop. There were 4 pieces that did more rotations than horizontal movements, so I was able to start holding Down 2 frames earlier on each of those.
And armed with this knowledge that spamming rotations can be done without losing frames, I'm going to try to abuse it. Rotating a piece close to a wall can cause some odd reorientations. Observe, I'm holding down here:
And now press A:
And press A again to get it where it was before. This managed to save 4 frames by reducing two Left inputs.
I'm not sure where else this can be done; I tested it on this piece due to having the tall I-piece wall, which is a pretty rare scenario to have. Probably at least 80% of the pieces dropped have no way of shaving directional input like this. I'll check as thoroughly as I can tonight.
EDIT:
Things might get more broken now. As I can go from this:
To this, using two simultaneous rotations:
If I didn't do it simultaneously, the piece would get locked in the other orientation. So this seems like a promising option for S and Z pieces.
It's funny, I spent quite some time trying to optimize WHERE I place the pieces, but didn't give enough effort into HOW I place the pieces, haha.
This chart may help understand how these rotation clips work: https://tetris.wiki/Tetris_(NES,_Tengen)#Rotation_rules
(I'd use a url tag but it wipes the whole message?)
My takeaway from it is that, as the rotation boxes are all pushed into the leftmost position (except I), that rotation clips like this might only be possible in a leftward direction? I'll trial and error as many possible placements as I can, though.
Alright, I found 42 different RNG options, about half of which lose 0 frames and half lose 1 frame. Expanding to 2 frames would immensely increase the options of course.
Also, most if not all of the 0 frame losses have Silence for the music; in my testing, you can do an Up or Down with the Start to choose Loginska or Troika at the cost of 1 lag frame, without changing the RNG. Would definitely want to do that for entertainment.
There's so many different ways you can input in the menus, and often, taking 2 or 3 frames on one menu will just reduce a lag frame in another menu, breaking even. And with both controllers active, you can do a lot of crazy things. For instance, to go Up to max level and max garbage, you can use Select or Down if you have 2 frames to play with:
And that'll get you different RNG, presumably because of different amount of cycles used.
Here's the BizHawk 2.4 tasproj that has a branch for each of these 42 options (and I'm sure I missed some): https://drive.google.com/open?id=1LOf4A6-cuLEWE1f1rbochfz2lyUF93N1
I have faith that at least one of these can do a clean 8 line clear run. Then the question would be, which can do an 8 line clear the fastest... Which is not my expertise, optimizing Tetris input seems very difficult.
Added an I-piece drought list to the Lua script, as that's probably the most crucial indicator of if a given RNG can possibly do an 8 line clear.
trying some of the options now, prioritizing ones with only one missing block in the top row of their garbage
41/218 (branch 35) likely failure due to massive 23 piece I-drought at the start
155/97 (branch 34) 150 frame faster input end; 9 line clears; but a lot of keeping the stack extremely high; might be worth keeping and optimizing further
This RNG went very well; keeping the stack high like that and using the 18 free blocks in the garbage is almost definitely more important than one less line clear animation...
(I have improved it by 11 frames since, by delaying the last Tetris a couple moves to have more height)
Well, I submitted that. that took less RNG digging than I expected
http://tasvideos.org/6635S.html
With this recent 30 lines submission, it's made trying to optimize PikachuMan's run sound like a good idea. Spikestuff started it, and I was interested and joined in.
It seems the most important way to save time in 30 lines is by completing it in only 8 line clears (7 tetrises and a double, or 6 tetrises and two triples), as the line clear animation is the same length regardless of how many lines are cleared.
But due to the fixed piece RNG upon game start, it's hard to say if such a possibility exists, especially if the RNG gives too long of an I-piece drought. I had a decent attempt going, probably would have been 9 line clears, but had a horrid drought with a ton of Z-blocks: http://tasvideos.org/userfiles/info/61514363070078918
Of course, other ways to save time are to play as high on the board as possible, and reduce left/right movement, but the piece order would eliminate most options for that. If you have a good layout to the starting handicap garbage, they could also be used (especially at the end) to save a few piece drops.
I'm working on a Lua script that displays extra upcoming pieces (it'll start by simulating the game to check the next 80 pieces; note this will overwrite Branch 0 in tastudio). Should help a lot with planning moves.
https://pastebin.com/4H75aayN
(Spikestuff came up with a nice design for more upcoming pieces, and also showing a red line for the play area boundary)
Currently creating a BizHawk 2.4 tasproj that has a branch for every possible starting RNG I can find. There are a surprising number of them that lose 0 or 1 frames; it must be cycle-based.
Well, rather than a disassembly approach, I decided to take a brute force approach and simply simulate what the upcoming 80 blocks will be.
Here's a Lua script that will show you 7 extra upcoming pieces (designed for 2x window size)
https://pastebin.com/4H75aayN
Although technically, this is off-topic, so I'll move any further optimization talk to the game thread for Tengen Tetris: http://tasvideos.org/forum/viewtopic.php?t=3761
As for Fraplatek's (two) submission(s), well, the main time save in this game is minimizing the number of separate line clears, and this run certainly doesn't do that.
I watched the old Tengen Tetris TAS just yesterday or so, oddly enough. And I did notice that it seemed like it had a low amount of Tetrises, since the animation for a line clear is the same length no matter how many lines.
Spikestuff offered me to help him try and get a high-Tetris/Triple run after I expressed interest, and it sounded fun enough. At the moment, with a very horrible ending, I have 1604 frames (26.689 seconds, vs. this one's 28.919): http://tasvideos.org/userfiles/info/61514363070078918
So obviously this movie isn't optimized, and neither is mine, but I think it would be interesting if a little frame war starts here... Anyone else interested?
This feels like there may be a more Lua-centric way to figure out the best possible starting RNG and setup, but I wouldn't even know where to begin with that.
Something that would be very useful is a Lua script that shows the next few pieces, however.
So far I've found:
0034 and 0035 seem to seed the RNG (change every frame)
005C and 005D are the in-game RNG (only change when a new piece is pulled)
Spikestuff found 0066 is the next piece:
1 - I
2 - T
3 - O
4 - J
5 - L
6 - S
7 - Z
Trying to debug, instruction 9930 is where 0066 is written, and 9A08 seems to be where 005C is written afterwards.
I just realized an improvement, as I wasn't totally happy with how technically useless the "rotate A leftwards" is and double-checking if it was necessary.
Turns out I did overlook that "_" is an option for opcode 57 in the lowercase alphabet, and it means I don't have to rotate A leftwards anymore to use 0F 57. It saved 2 frames. It also makes the payload 8 characters, which saves 2 frames (since I can press B one less time at the start). I definitely can't see any way to compress the payload any more than that now.
Here's the file: http://tasvideos.org/userfiles/info/61405334239396869
I'm not sure if you'd prefer "GR Dan Sanjou!" in the game name; googling Pokemon Card GB2 suggests that "Here Comes Team GR!" has become the most common subtitle (and is used on Bulbapedia and Speedrun.com).
We always use the romanised title when the game was never released in the USA, not the translated title. There's at least one precedent.
Does this apply to bootlegs also? As a fan of Pokemon TASes specifically, it's frustrating to see "Platinum Edition" smack dab in the middle of this page where its placement (right after Pearl, by coincidence) makes it much more easily confused with Platinum Version. If we're using romanized titles instead, the bootleg's title screen says 口袋妖怪 白金版 and the romanization can be found in the submission's ROM filename.
Sorry for the slight off-topic.
That's an interesting example. Especially since "Pokemon Platinum" is clearly a much worse search term to find the game than "Kou Dai Yao Guai - Bai Jin Ban".
Using translated titles is definitely tricky because they can be translated differently by different people. Heck, the English translation we all use has a title screen of "Pokemon Trading Card Game 2: The Invasion of Team GR!" so that's different than what everyone else seems to be using. Although the romanization could be "GR Dan Sanjou!" or "GR-dan Sanjou!"
For bootlegs, I guess it makes sense their English box art name would be used, as its just as official a release as the country it originated from. I cannot personally find any picture of a box or cartridge for this Pokemon Platinum Edition. But the use of the word "Edition" instead of "Version" is curious, and makes me assume it did originate from the name on an 'official' release outside of China.
So I do think commonly-accepted English branding of bootlegs is okay, as confusing as it may be in that case. And I do think "GR-dan Sanjou!" is the best name for this submission. (Going by the Giant Bomb wiki, like Twitch does.)
It seems playing as the girl wastes (at least) 1 frame. Does doing this actually make the setup faster?
Nah, the gender selection is done with a Right+A input on the same frame, and "boy" and "girl" are the same number of Japanese characters ("boy" would be faster in the US fan translation)
So gender and name don't lose time (as long as the name is input in one frame, so you could do A, Left+A, Down+A, or Right+A)
And now press A:
And press A again to get it where it was before. This managed to save 4 frames by reducing two Left inputs.
I'm not sure where else this can be done; I tested it on this piece due to having the tall I-piece wall, which is a pretty rare scenario to have. Probably at least 80% of the pieces dropped have no way of shaving directional input like this. I'll check as thoroughly as I can tonight.
EDIT:
Things might get more broken now. As I can go from this:
To this, using two simultaneous rotations:
If I didn't do it simultaneously, the piece would get locked in the other orientation. So this seems like a promising option for S and Z pieces.
It's funny, I spent quite some time trying to optimize WHERE I place the pieces, but didn't give enough effort into HOW I place the pieces, haha.
This chart may help understand how these rotation clips work: https://tetris.wiki/Tetris_(NES,_Tengen)#Rotation_rules
(I'd use a url tag but it wipes the whole message?)
My takeaway from it is that, as the rotation boxes are all pushed into the leftmost position (except I), that rotation clips like this might only be possible in a leftward direction? I'll trial and error as many possible placements as I can, though.
And that'll get you different RNG, presumably because of different amount of cycles used.
Here's the BizHawk 2.4 tasproj that has a branch for each of these 42 options (and I'm sure I missed some): https://drive.google.com/open?id=1LOf4A6-cuLEWE1f1rbochfz2lyUF93N1
I have faith that at least one of these can do a clean 8 line clear run. Then the question would be, which can do an 8 line clear the fastest... Which is not my expertise, optimizing Tetris input seems very difficult.
Added an I-piece drought list to the Lua script, as that's probably the most crucial indicator of if a given RNG can possibly do an 8 line clear.
trying some of the options now, prioritizing ones with only one missing block in the top row of their garbage
41/218 (branch 35) likely failure due to massive 23 piece I-drought at the start
155/97 (branch 34) 150 frame faster input end; 9 line clears; but a lot of keeping the stack extremely high; might be worth keeping and optimizing further
This RNG went very well; keeping the stack high like that and using the 18 free blocks in the garbage is almost definitely more important than one less line clear animation...
(I have improved it by 11 frames since, by delaying the last Tetris a couple moves to have more height)
Well, I submitted that. that took less RNG digging than I expected
http://tasvideos.org/6635S.html
Currently creating a BizHawk 2.4 tasproj that has a branch for every possible starting RNG I can find. There are a surprising number of them that lose 0 or 1 frames; it must be cycle-based.